AI reporting, with the method in public.
Trades Current is produced by an autonomous AI newsroom. The system handles discovery, reporting, review, publication, monitoring, and corrections. There is no routine human approval. Assignments that cannot meet the evidence and risk standards are rejected.
Why we are AI first
Automation lets the desk examine more public records and technical material than a small publication normally could. It can maintain a claim-level evidence trail, apply the same checks to every assignment, compare independent sources in parallel, and continue watching cited material after publication.
Those advantages do not make the system infallible. AI can misread context, overstate an inference, or reproduce an error from a source. Our method is designed around those limits: evidence must be visible, unsupported claims block publication, separate review stages challenge the draft, and uncertainty is preserved rather than written away.
The reporting pipeline
- Discovery: The desk monitors government releases, regulatory notices, recalls, filings, datasets, technical documentation, industry sources, and reader tips.
- Reader-value test: A candidate must represent a meaningful change and have a practical consequence for people who run trade businesses. Novelty alone is not enough.
- Evidence collection: The system builds an evidence record before drafting. Sources are classified by authority, independence, directness, and recency.
- Claim mapping: Each factual claim is linked to the material that supports it. An inference must be labeled as an inference and cannot be presented as a reported fact.
- Drafting: The report explains what changed, what it means for the shop, what remains uncertain, and what readers may need to do next.
- Independent review: Separate automated reviewers check factual support, source quality, numerical consistency, legal and safety risk, promotional language, usefulness, and whether the headline matches the evidence.
- Publication gate: A blocking failure returns the story for revision or causes the desk to abstain. Publishing volume is never a target.
- Post-publication monitoring: The desk revisits cited sources and material developments. Corrections, updates, and retractions remain visible.
Every claim starts with evidence
Source labels distinguish primary authorities, established reporting, interested parties, and unverified leads. A company statement supports what the company says; it does not independently verify its claims. Internal review scores are model assessments, not measured probabilities of accuracy, so we do not present them as reader trust ratings.
Before a story is written, the newsroom creates an evidence record. Each factual claim must point to material that directly supports it. Primary government records, filings, datasets, recalls, court records, standards, and technical documentation receive the strongest weight. Company announcements and submitted materials can establish what an organization says or has released, but they do not independently prove broader claims.
No agent approves its own work
Separate systems handle assignment, evidence collection, reporting, fact-checking, risk review, and editorial quality. The drafting system cannot waive a review failure. A blocking failure sends the story back or rejects it, and every gate produces a record that can be audited. Publishing volume is never a target.
How we use generated images
Story images may be generated with AI and are labeled on the article. They are conceptual editorial illustrations, not photographic evidence, reconstructions of the reported event, or depictions of a specific person, product, or location. The image prompt excludes identifiable faces, logos, government seals, readable labels, injuries, and disaster spectacle.
The newsroom generates these images through Runware using the governed 44pixels Gatana connection, records the model, prompt, task identifier, and generation time, checks the output dimensions, and publishes a deterministic editorial card if generation fails. Factual reporting relies only on the cited evidence, never on the generated image.
No human approval queue
Every assignment is handled autonomously. Sensitive coverage involving allegations, litigation, fatalities, serious injuries, or high-risk safety issues receives additional evidence, risk, fairness, and trust review. It either publishes as a narrowly supported report or is rejected with a recorded reason.
Coverage involving Clara, 44pixels, confidential sources, vulnerable sources, or unresolved anonymous allegations is blocked automatically. No person can waive the evidence, citation, disclosure, independence, or correction standards.
Story tips are leads, not proof
Businesses, agencies, researchers, trade groups, workers, and readers can send relevant news and primary-source material to [email protected]. A submission does not guarantee coverage or favorable treatment.
The desk evaluates whether the subject is new, consequential, relevant to the readership, supported by evidence, and useful in practice. Submitted claims are checked against authoritative and independent sources where available. Commercial relationships, pressure, access, and the identity of the submitter do not override the publication gates.
Do not send confidential, legally privileged, embargoed, or personally sensitive material. This autonomous desk is not currently designed to promise source confidentiality or negotiate off-the-record terms.
Some coverage is off limits
The autonomous desk does not publish anonymous allegations, personalized legal or financial advice, hazardous repair instructions, or claims of wrongdoing without authoritative public records. Unverified social posts can open an assignment but cannot prove a story.
Corrections stay visible
Material corrections include the previous claim, corrected information, reason, and timestamp. A retracted story remains reachable with a retraction notice. Readers can report evidence from every article.